15
Sensors 2017, 17, 2502; doi:10.3390/s17112502 www.mdpi.com/journal/sensors Article COALA: A Protocol for the Avoidance and Alleviation of Congestion in Wireless Sensor Networks Dionisis Kandris 1, *, George Tselikis 2 , Eleftherios Anastasiadis 3 , Emmanouil Panaousis 4 and Tasos Dagiuklas 3 1 Department of Electronic Engineering, Technological Educational Institute of Athens, 12243 Athens, Greece 2 Department of Computer Systems Engineering, Piraeus University of Applied Sciences, 12241 Athens, Greece; [email protected] 3 School of Engineering/Computer Science and Informatics, London South Bank University, SE1 0AA London, UK; [email protected] (E.A.); [email protected] (T.D.) 4 Department of Computer Science, University of Surrey, GU2 7XH Guildford, UK; [email protected] * Correspondence: [email protected]; Tel.: +306972256162 Received: 29 September 2017; Accepted: 28 October 2017; Published: date Abstract: The occurrence of congestion has an extremely deleterious impact on the performance of Wireless Sensor Networks (WSNs). This article presents a novel protocol, named COALA (COngestion ALleviation and Avoidance), which aims to act both proactively, in order to avoid the creation of congestion in WSNs, and reactively, so as to mitigate the diffusion of upcoming congestion through alternative path routing. Its operation is based on the utilization of an accumulative cost function, which considers both static and dynamic metrics in order to send data through the paths that are less probable to be congested. COALA is validated through simulation tests, which exhibit its ability to achieve remarkable reduction of loss ratios, transmission delays and energy dissipation. Moreover, the appropriate adjustment of the weighting of the accumulative cost function enables the algorithm to adapt to the performance criteria of individual case scenarios. Keywords: Wireless Sensor Networks; congestion avoidance; congestion control; load balancing; energy efficiency; routing protocol 1. Introduction The operation of a WSN is interdependently correlated with the existence of data traffic. Network congestion is one of the most serious problems encountered in the management of the data traffic within a WSN. Congestion occurs when current traffic load exceeds available transmission ability at any point in the network. Congestion has an absolutely detrimental impact on WSN performance [1]. More specifically, congestion procures the overflow of node buffers, the degradation of the overall channel quality, and the increase of both loss rates and transmission delays. This article proposes a novel protocol that aims, by ruling the routing process, not only to prevent the occurrence of congestion, but also to deter the dispersion of oncoming congestion in WSNs. This protocol performs the discovery of the routing paths that are less likely to be congested, based on the computation of a suitably formulated cost function. This cost function takes into consideration a collection of static and dynamic metrics that are related with congestion. The remainder of this article is organized as follows. Section 2 outlines various existing protocols for congestion avoidance and in congestion control WSNs. The description of the

Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

  • Upload
    others

  • View
    7

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

   

Sensors  2017,  17,  2502;  doi:10.3390/s17112502   www.mdpi.com/journal/sensors  

Article  

COALA:  A  Protocol  for  the  Avoidance  and  Alleviation  of  Congestion  in  Wireless  Sensor  Networks  Dionisis  Kandris  1,*,  George  Tselikis  2,  Eleftherios  Anastasiadis  3,  Emmanouil  Panaousis  4  and  Tasos  Dagiuklas  3  1   Department  of  Electronic  Engineering,  Technological  Educational  Institute  of  Athens,  12243  Athens,  

Greece  2   Department  of  Computer  Systems  Engineering,  Piraeus  University  of  Applied  Sciences,  12241  Athens,  

Greece;  [email protected]  3   School  of  Engineering/Computer  Science  and  Informatics,  London  South  Bank  University,  SE1  0AA  

London,  UK;  [email protected]  (E.A.);  [email protected]  (T.D.)  4   Department  of  Computer  Science,  University  of  Surrey,  GU2  7XH  Guildford,  UK;  

[email protected]  *   Correspondence:  [email protected];  Tel.:  +30-­‐‑697-­‐‑225-­‐‑6162  

Received:  29  September  2017;  Accepted:  28  October  2017;  Published:  date  

Abstract:  The  occurrence  of  congestion  has  an  extremely  deleterious  impact  on  the  performance  of  Wireless   Sensor   Networks   (WSNs).   This   article   presents   a   novel   protocol,   named   COALA  (COngestion  ALleviation  and  Avoidance),  which  aims   to  act  both  proactively,   in  order   to  avoid   the  creation   of   congestion   in   WSNs,   and   reactively,   so   as   to   mitigate   the   diffusion   of   upcoming  congestion   through   alternative   path   routing.   Its   operation   is   based   on   the   utilization   of   an  accumulative  cost  function,  which  considers  both  static  and  dynamic  metrics  in  order  to  send  data  through  the  paths  that  are  less  probable  to  be  congested.  COALA  is  validated  through  simulation  tests,  which  exhibit   its  ability  to  achieve  remarkable  reduction  of   loss  ratios,   transmission  delays  and   energy   dissipation.   Moreover,   the   appropriate   adjustment   of   the   weighting   of   the  accumulative  cost  function  enables  the  algorithm  to  adapt  to  the  performance  criteria  of  individual  case  scenarios.  

Keywords:  Wireless  Sensor  Networks;  congestion  avoidance;  congestion  control;   load  balancing;  energy  efficiency;  routing  protocol  

 

1.  Introduction  

The   operation   of   a   WSN   is   interdependently   correlated   with   the   existence   of   data   traffic.  Network   congestion   is   one   of   the  most   serious   problems   encountered   in   the  management   of   the  data   traffic   within   a   WSN.   Congestion   occurs   when   current   traffic   load   exceeds   available  transmission  ability  at  any  point  in  the  network.  Congestion  has  an  absolutely  detrimental  impact  on  WSN  performance  [1].  More  specifically,  congestion  procures  the  overflow  of  node  buffers,  the  degradation   of   the   overall   channel   quality,   and   the   increase   of   both   loss   rates   and   transmission  delays.  

This   article   proposes   a   novel   protocol   that   aims,   by   ruling   the   routing   process,   not   only   to  prevent   the  occurrence  of   congestion,   but   also   to  deter   the  dispersion  of   oncoming   congestion   in  WSNs.  This  protocol  performs  the  discovery  of  the  routing  paths  that  are  less  likely  to  be  congested,  based   on   the   computation   of   a   suitably   formulated   cost   function.   This   cost   function   takes   into  consideration  a  collection  of  static  and  dynamic  metrics  that  are  related  with  congestion.  

The   remainder   of   this   article   is   organized   as   follows.   Section   2   outlines   various   existing  protocols   for   congestion   avoidance   and   in   congestion   control   WSNs.   The   description   of   the  

Page 2: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     2  of  15  

proposed  congestion  protocol  takes  place  in  Section  3.  In  Section  4,  the  performance  evaluation  of  the   proposed   protocol   is   performed   through   the   description   and   analysis   of   simulation   results.  Finally,  Section  5  concludes  the  article.  

2.  Related  Work  

Congestion   is   a   phenomenon   that   comes   along   either   interference   in   the   communication  medium,  which  is  caused  by  the  concurrent  transmission  of  many  nodes,  or  buffer  overflow,  which  is   caused   by   the   fact   that   incoming   traffic   load   in   a   node   exceeds   its   buffer   capacity.   The  confrontation  of  congestion  is  the  subject  of  numerous  scientific  research  works  [1–7].  Some  of  them,  introduce   congestion   avoidance   protocols,   which   propose   proactive   tasks   in   order   to   prevent  congestion  occurrence.  These  protocols  normally  involve  MAC  and  network  layer  operations.  Some  other   research   works   propose   congestion   control   protocols   that   act   reactively   to   the   existence   of  congestion  in  order  to  mitigate  it.  Protocols  of  this  kind  normally  involve  MAC  and  network  layer  operations,   and   in   some   cases   they   also   use   transport   layer   actions.   Additionally,   cross   layer  interaction  between  transport  and  underlying  layers  is  an  efficient  way  of  congestion  control  while  MAC  layer  provides  channel  status  that  can  be  incorporated  in  congestion  control  mechanisms  [1].  

The   detection   of   congestion   can   be   performed   by   taking   into   consideration   one   or   a  combination   of   specific   performance   metrics.   The   most   popular   of   them   are:   packet   loss,   buffer  occupancy,  delay,  packet  service  time  and  packet  inter-­‐‑arrival  time  [2,3].  

The   notification   of   congestion   occurrence   can   be   either   explicit,   where   relative   informing  messages  are  sent  by  congested  nodes  to  other  nodes,  or  implicit,  where  the  notifying  information  is  incorporated  in  data  packet  headers  or  in  ACK  packets  that  are  piggybacked.  Explicit  notification  is  unfavorable  because  it  adds  substantial  traffic  load  to  the  already  jammed  network  [4,5].  

The  mitigation  of  congestion  can  be  pursued  by  either   traffic  control  or  resource  control  or  a  combination  of   them.  When   traffic   control   is   applied,   the   quantity   of   the  packets   injected   into   the  network  is  suitably  decreased  in  order  to  alleviate  both  traffic  load  and  congestion.  However,  traffic  control  is  not  efficient  in  event-­‐‑based  applications  where  any  restriction  in  the  transmission  of  data  is  inacceptable.  When  resource  control   is  applied,  data  packets  are  routed  through  alternative  paths  that  are  not  congested.  Yet,  in  this  way  extra  delays  or  even  routing  loops  may  be  caused  [6,7].  

CODA,  proposed  by  Wan  et  al.  [8],  is  one  of  the  most  well  -­‐‑known  protocols,  which  aims  at  the  achievement   of   congestion   avoidance   and   control.   Its   operation   is   based   on   flow   control.   It  introduces   the   idea   of   a   control  mechanism   by  which   every   node   that   detects   the   occurrence   of  congestion,  sends  backpressure  messages  to  its  data  source  nodes.  Every  source  node  that  receives  backpressure  signals  either  throttles  its  sending  rates  or  drops  packets  based  on  the  local  congestion  policy  adopted.  Additionally,   source  nodes   start   throttling   their   sending   rates  as   soon  as   they  do  not  receive,  at  predefined  time,  feedback  messages  sent  to  them  by  sink.  

Ahmad  and  Turgut  [9]  proposed  an  alternative  path  routing  protocol  which  uses  the  ratio  of  the   numbers   of   downstream   to   upstream   nodes   along   with   the   queue   sizes   of   the   downstream  nodes  in  order  to  detect  congestion  and  reallocate  traffic  through  alternate  routes.  

PACA,  proposed  by  Kandris  et  al.  [10],  pursuits  congestion  avoidance  by  circumventing  nodes  that  are  either  located  close  to  the  sink,  or  have  more  downstream  than  upstream  neighbors  or  have  very  frequent  data  transmissions.  

In   [11]   Sergiou   et   al.   introduce   two   very   promising   lightweight   schemes   for   congestion  avoidance  and  control.   In  the  first  of   them,  named  DAlPaS  Hard,  data  flows  are  forced  to  change  their  path  in  order  not  to  congest  the  receiving  node  based  on  a  multivariable  utility  function.  In  the  second  of  them,  named  DAlPaS,  each  node  attempts  to  serve  just  one  flow  and  if  this  is  unavoidable  the  DAlPaS  Hard  algorithm  is  executed.  

Huang  et  al.  proposed  TALONet  [12],  which  uses  both  traffic  control  and  resource  control  to  avoid  congestion.  Specifically,  TALONet  combines  various  levels  of  transmission  power  in  order  to  relieve  existing  congestion  in  data  link  layer,  along  with  buffer  management  to  avoid  congestion  in  buffer   level,   and  multi-­‐‑path   routing   in   order   to   relay   congested   traffic   flows   through   alternative  paths.  

Page 3: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     3  of  15  

HTAP   [13],  proposed  by  Sergiou  et   al.,   is   a   scalable  protocol   aiming   to  minimize   congestion  and  assure  reliable  data  transmissions  in  event-­‐‑based  networks  through  resource  control.  As  soon  as   congestion   occurrence   is   detected,   alternative   paths   are   created   and   nodes   are   hierarchically  levelled  in  these  paths  and  exhausted  nodes  are  bypassed  in  order  to  achieve  load  balancing.  

He  et  al.  proposed  TADR  [14]  which  uses  shortest  paths  in  order  to  route  data  when  there  is  no  congestion.  As  soon  as  congestion  is  detected,  the  concept  of  potential  fields  is  used,  by  taking  into  consideration   network   depth   and   normalized   queue   length,   in   order   to   distribute   data   through  multiple  paths  consisting  of  idle  and  under-­‐‑loaded  nodes  so  as  to  circumvent  congested  areas.  

Tao   and  Yu   introduced  ECODA   [15],  where  packets   are  dynamically  prioritized,  using   their  initial   static   packet   priority,   hop-­‐‑count   and   the   time   from   the   packet   generation   to   current   time.  ECODA   proposes   buffer   related   metrics   so   as   to   detect   congestion.   The   congestion   status   is  piggybacked   in   packets.   When   receiving   such   a   back-­‐‑pressure   message,   the   source   node   either  reduces   its   transmission   rate,   or   accordingly   adjusts   the   rate   for   different   paths   if  multiple   paths  exist.  

Kang   et   al.   proposed   the   TARA   protocol   [16].   In   TARA,   as   soon   as   emerging   congestion   is  detected   in   a   node,   by  measuring   both   the   buffer   occupancy   and   the   channel   load,   this   node   is  considered   to   be   a   hot-­‐‑spot   node.   Next,   traffic   is   deflected   from   the   hot-­‐‑spot   node   through   a  so-­‐‑called   distributor   node   along   a   detour   path   and   reaches   the   so-­‐‑called  merge   node,  where   the  flows  are  merged.  As   soon  as   congestion  has  been  alleviated   the  network   stops  using   the  detour  path.  

Jan  et  al.  introduced  PASCCC  [17],  which  is  an  energy-­‐‑efficient  application  specific  clustering  congestion   control   protocol.   In   PASCCC,   data   packets   are   prioritized   as   high   priority   and   low  priority  packets  according  to  the  type  of  their  content.  During  congestion  low  priority  packets  are  discarded.  

Table  1  enlists  in  comparison  basic  characteristics  of  both  the  aforementioned  protocols  and  the  proposed  in  this  research  work  protocol.  

Table  1.  Characteristic  features  of  congestion  avoidance  and  control  protocols.  

Protocol   Metric  Considered  Congestion  Notification  

Congestion  Mitigation/Avoidance  

CODA  [7]   Buffer  occupancy  and  channel  load   Explicit   Traffic  Control  Ahmad  and  Turgut  [8]  

Buffer  occupancy  and  characteristic  ratio   Implicit   Traffic  Control  

PACA  [9]  Buffer  occupancy,  characteristic  ratio,  

distance,  time  of  use  Implicit   Resource  Control  

DAlPaS  [10]   Buffer  occupancy,  channel  load,  energy   Implicit   Resource  Control  TALONet  [11]   Buffer  occupancy   Implicit   Traffic  and  Resource  Control  HTAP  [12]   Buffer  occupancy,  energy   Implicit   Resource  Control  TADR  [13]   Buffer  occupancy   Implicit   Resource  Control  

ECODA  [14]  Dual  buffer  threshold  and  weighted  

buffer  difference  Implicit   Traffic  Control  

TARA  [15]   Buffer  occupancy  and  channel  load   Explicit   Resource  Control  PASCCC  [16]   Buffer  occupancy,  type  of  content   Implicit   Traffic  Control  

COALA  Buffer  occupancy,  popularity  index,  

energy,  distance,  vicinity  index  Implicit   Resource  Control  

3.  Proposed  Protocol  Description  

Following  the  above-­‐‑mentioned  introduction  to  well-­‐‑known  protocols  for  the  avoidance  and  or  control   of   congestion   in   WSNs,   COALA   (COngestion   ALleviation   and   Avoidance),   which   is   the  proposed  in  this  article  protocol  of  this  kind,  is  introduced  in  this  section.  

COALA  protocol,   operating   proactively,   uses   simple   static   information   of   network   nodes   in  order  to  perform  data  routing  through  the  paths  that  are  less  probable  to  be  congested.  In  the  face  

Page 4: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     4  of  15  

of   imminent   congestion,   COALA   acting   reactively,   uses   implicit   notification   of   congestion   and  applies  resource  control  with  the  intention  of  preventing  the  further  diffusion  of  congestion.    

Page 5: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     5  of  15  

3.1.  Preliminary  Considerations  and  Terms  

First,   a   densely   deployed   WSN   where   homogeneous   nodes   are   positioned   in   a   uniformly  random  way,  such  as  the  one  illustrated  in  Figure  1a,  is  considered  to  be  the  reference  point  for  the  protocol   description.   It   is   assumed   that   there   are   is   one   sink.   It   is   also   supposed   that   every  individual  node  knows  both  its  location  and  the  location  of  the  sink.  

   (a)   (b)  

Figure   1.   A   typical   example   of   a   WSN   topology   in:   (a)   Initial   arrangement   of   interconnected  network  nodes;  (b)  Level-­‐‑based  taxonomy  of  network  nodes.  

Additionally,  it  is  assumed  that  incidents  referred  here  as  events,  are  created  and  all  nodes  that  are  located  within  the  range  of  each  event,  send  related  data  to  the  sink.  Thus,  the  number  of  data  sources  is  variable.  Also,  all  nodes  are  considered  to  have  the  same  transmission  and  sensing  range.  Moreover,   it   is   supposed   that   multihop   routing   through   direct   neighboring   nodes   is   utilized,  whenever   the  sink   is  not  placed  within   the   transmission  range  of   the  node  that  relays  data   to   the  sink.  Furthermore,  CSMA  is  considered  to  be  used  as  the  medium  access  control  (MAC)  protocol.  

For  every  individual  node  i:  

• the  area  where  the  signals  sent  by  node  i  can  reach,  is  defined  as  Transmission  Range  TR(i)  • the  minimum  number  of  hops  for  node  i  in  order  to  reach  the  sink,  is  defined  as  level  L(i)  • every  node   j   that   is  placed  within   tr(i)   and  has  L(j)  =  L(i)  ±  1,   is  defined  as  Neighboring  Node  

NN(i)  • the  ratio  of  the  total  number  of  NN(i)  whose  level  L  is  greater  than  L(i)  to  the  total  number  of  

NN(i),  is  defined  as  Vicinity  Index  VI(i)  • the   ratio   of   the   accumulative   participation   of   a   node   in   data   flows   over   time   is   referred   as  

Popularity  Index  PI(i)  • the   so   called  Availability   Index  AI(i)   expresses   either   the   ability   (when  having  value   1)   of   the  

node  to  relay  data  or  the  unavailability  (when  having  value  0)  of  the  node  to  do  so,  because  the  node  has  either   limited  buffer  space,  or   limited  energy  or  because   its  neighboring  nodes  that  have  lesser  level  are  blocked  

3.2.  Initialization  Phase  

The  initialization  phase  of  COALA  aims  to  perform  all  the  initial  calculations  of  both  the  level  L  and  the  neighbor  table  of  all  networks  nodes.  These  calculations  are  necessary  for  the  inception  of  the  congestion  avoidance  algorithm  and  this  is  why  it  is  executed  only  once.  

Specifically,   this   phase   is   initiated   as   soon   as   the   sink   transmits   a   “hello”   message,   which  includes   the   sink   ID   number   along   with   its   level   L,   which   is   equal   to   0.   Next,   every   node   that  receives   this   message,   sends   back   to   the   sink   a   corresponding   acknowledgement   message   that  

Page 6: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     6  of  15  

includes   its   ID   number.  As   soon   as   the   sink   receives   such   an   acknowledgement  message   from   a  node,   it   sends  another  message  back   to   this  node,  which  confirms  that   this  node   is  a  neighboring  node  of  the  sink  and  that  it  is  a  level  1  node.  Every  level  1  node  initiates  its  neighbor  table,  which  includes  the  details  of  the  sink  and  transmits  a  “hello”  message,  which  includes  its  ID  number,  its  level  L,  which  is  equal  to  1,  its  position,  its  current  energy,  and  its  current  buffer  occupancy.  Every  node   that   receives   this   message,   sends   back   to   the   transmitting   node   a   corresponding  acknowledgement  message  that  includes  the  current  values  of  its  neighbor  table  parameters,  which  are   explained   later   on.   As   soon   as   the   level   1   node   receives   such   an   acknowledgement  message  from  a  node,   it   sends  another  message  back   to   this  node   that   includes   its  updated  neighbor   table  along  with  a  confirmation  that  this  node  is  its  neighboring  node.  If  the  new  neighboring  node  has  not  already  acquired  a  level  number,  then  it  is  recognized  as  a  next  (i.e.,  2)  level  node.  

This   procedure   carries   on   until   every   individual   node   in   the   network   not   only   has   been  assigned  a  corresponding  level  number,  as  illustrated  in  Figure  1b,  but  also  has  become  aware  of  its  neighbor   table.   In   this   way,   each   node   constructs   an   overall   view   of   the   network   topology   and  becomes   aware   of   all   the   available   routing   paths   towards   the   sink,   avoiding   the   formation   of  routing  loops.  

3.3.  Steady-­‐‑State  Phase  

The   steady-­‐‑state,   which   is   explained   in   this   subsection,   begins   as   soon   as   the   initialization  phase  is  completed.  Its  operation  is  mainly  based  on  the  utilization  of   the  neighbor  tables  that,  as  mentioned   above,   have   been   created,   during   the   initialization   phase,   for   all   network   nodes.   A  typical  neighbor  table  of  a  network  node  contains  the  current  values  of  the  following  parameters:  

• node  ID,  • level  number  • position  • energy  • buffer  occupancy  • popularity  index  • availability  index  • vicinity  index  

The  overall  data  routing  process,  during  the  steady-­‐‑state,  is  based  on  the  current  values  of  all  neighbor   tables   that   are   dynamically   updated.   Specifically,   similarly   to   [11]   every   network  node,  which  has  data   to   send   to   the   sink,   examines   its  own  neighbor   table  and  discovers   the  candidate  recipients  among  its  neighbors  who  have  smaller  level  number,  i.e.,  its  neighbors,  which  are  located  at  a  level  closer  to  the  sink,  if  any.  

Therefore,  at  this  point,  each  node  neglects  all  of   its  connections  with  other  nodes,  which  are  located  either  at  the  same  level  with  itself  or  at  lower  levels  and  focuses  at  the  nodes  that  are  placed  at  an  upper  level  in  the  network  taxonomy,  as  shown  in  Figure  1b.  

In  the  case  that  there  are  at  least  two  candidate  recipient  nodes  having  AI  =  1,  COALA  protocol  suggests   that   the   transmitting   node  must   relay   its   data   through   the   node   that   has   the  minimum  vicinity  index.  For  instance,  referring  to  Figure  1b,  if  node  14  has  data  to  send  upwards  to  the  sink,  then  it  has  two  candidate  nodes  through  which  it  can  relay  its  data,  i.e.,  node  9  and  node  10.  If  these  nodes  are  both  available  (i.e.,  have  AI  =  1)  then  node  9  will  be  selected,  since  VI(9)  =  2/5  =  0.4  and  VI(10)  =  3/4  =  0.75.  This  is  because  node  9  may  receive  data  from  2  lower  level  nodes  and  transmit  data  to  3  upper  level  nodes,  while  node  10  may  receive  data  from  3  lower  level  nodes  and  transmit  data  to  1  upper  level  node.  Thus,  theoretically  node  10  is  more  likely  to  be  congested  than  node  9  if  normalized   flow   rate   patterns   are   considered.   This   initial   routing   criterion   enhances   the  corresponding  initial  consideration  suggested  in  [11].  

This  procedure   is  carried  on  until   the  whole  routing  path  has  been  determined.  The  selected  path  information  is  stored  in  the  header  of  the  data  packets  transmitted.  In  this  way,  every  time  that  a   node   has   data   to   send   to   the   sink,   a   dynamically   updated   spanning   tree   is   created   in   order   to  

Page 7: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     7  of  15  

route   all   data   load   through   the   theoretically   less   probable   to   be   congested   shortest   path.   It   is  important  to  notice  that  this  process  is  based  on  purely  static  information,  which  is  well-­‐‑known  as  soon  as  the  steady-­‐‑phase  is  terminated.  

However,   in   the   case   that   the   node   that   is   selected   to   be   the   next   to   receive   data,   becomes  unavailable   (due   to  either  buffer  or  energy   limitations  or   lack  of  available  upper   level  neighbors)  the  routing  process  necessarily  changes.  

Specifically,   the   data   has   to   be   relayed   via   an   alternative   path.   For   this   reason,   COALA  protocol  proposes   the  utilization  of  a  multivariable  cost   function  that  aims  to  evaluate   the  overall  cost  of  every  neighboring  node,  by   taking   into  consideration   the  buffer  occupancy,   the  dissipated  energy,   the   level   number,   the   availability   index,   the   popularity   index,   the   vicinity   index   and   the  geographical  distance  of  this  node.  The  mathematical  representation  of  this  cost  function  is  defined  in  (1):  

𝐴𝐶 𝑖 = 𝑤!" ∙ 𝐵𝑂 𝑖 + 𝑤!" ∙ 𝐷𝐸 𝑖 + 𝑤! ∙ 𝐿 𝑖 + 𝑤!" ∙ 𝑉𝐼 𝑖 + 𝑤!" ∙ 𝑃𝐼 𝑖 + 𝑤! ∙ 𝐷 𝑖 ∙ 𝐴𝐼 𝑖   (1)  

where:  

𝐴𝐶 𝑖  denotes  the  accumulative  cost  of  node  i  𝐵𝑂 𝑖   denotes  the  normalized  buffer  occupancy  of  node  i  𝐷𝐸 𝑖   denotes  the  normalized  dissipated  energy  of  node  i  𝐿 𝑖   denotes  the  normalized  level  number  of  node  i  𝑉𝐼 𝑖   denotes  the  normalized  vicinity  index  of  node  i  𝑃𝐼 𝑖   denotes  the  normalized  popularity  index  of  node  i  𝐷 𝑖   denotes  the  normalized  geographical  distance  of  node  i  from  the  transmitting  node  𝐴𝐼 𝑖   denotes  the  availability  index  of  node  i    𝑤!" ,𝑤!" ,𝑤! ,𝑤!" ,𝑤!" ,𝑤!     are  the  corresponding  weighting  factors  

Once  the  accumulative  cost  has  been  calculated  for  all  alternative  neighboring  nodes,  COALA  routing   algorithm   selects   the   node   that   has   the  minimum  value   of   accumulative   cost   as   the   next  recipient  of  the  data  in  the  routing  path  towards  the  sink.  

The  buffer  occupancy  metric  is  considered  in  order  to  prefer  nodes  that  have  more  free  space  in  their  buffers.  The  dissipated  energy  parameter  is  used  so  as  to  avoid  the  use  of  nodes  that  have  low  energy  reserves.  The  level  number  is  considered  in  order  to  give  priority  to  upper  level  nodes,  so   that  data   are   routed   towards   the   sink.   The  vicinity   index  metric   is   used   in   order   to   avoid   the  utilization   of   nodes   that   have   plenty   of   probable   data   suppliers   and   few   data   recipients.   The  popularity   index   is   taken   into   consideration   so   as   to   slide   over   network   nodes   that   tend   to   be  repetitively  busy.  For  instance,  nodes  that  are  located  either  in  centric  routing  paths  or  within  areas  with  frequent  creation  of  events  have  high  popularity  and  thus  are  more  probable  to  get  congested.  The  geographical  distance  between  neighboring  nodes  is  taken  under  consideration  in  order  to  give  increased  priority  to  the  data  relaying  via  the  closest  nodes  than  the  more  distant  ones.  The  overall  sum  of  products  is  multiplied  by  the  availability  index  in  order  to  prevent  the  utilization  of  nodes  that   are   unavailable   due   to   either   insufficient   energy,   or   inadequate   buffer   space   or   even  inaccessible  upper  level  neighboring  nodes.  

Additionally,   the  existence  of   the  weighting   factors   in   this  cost   function,  aims   to  support   the  appropriate   adaptation   of   the   algorithm   in   order   to   satisfy   the   different   demands   of   every  individual  application.  For   instance,  wherever  the  conservation  of  energy  has  major  priority,   𝑤!"  is  assigned  a  greater  value.  In  Figure  2,  a  descriptive  flowchart  of  COALA  protocol  is  illustrated.    

Page 8: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     8  of  15  

 Figure  2.  Flowchart  of  the  algorithm  of  COALA  protocol.  

4.  Proposed  Protocol  Evaluation  

The  proposed  congestion  avoidance  protocol  has  been  validated  through  the  utilization  of  an  appropriately   developed,   simulation   environment.   The   customized   console-­‐‑based   simulation  platform  was  built  by  using  C++  programming  language  according  to  the  methodologies  described  in  [18,19].  

Specifically,  multiple  runs  have  been  executed  in  order  to  investigate  the  protocol  performance  in   comparison   with   DAlPaS   Hard   scheme   in   various   scenarios   concerning   randomly   deployed  topologies.  

Page 9: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     9  of  15  

4.1.  Simulation  Process  

The  simulation  environment  creates  a  number  of  user  defined  nodes,  randomly  positioned  in  a  user  defined  geographical  area  of  square  shape.  The  node  that  is  created  first  plays  the  role  of  the  unique  sink.  The  user  defines  the  minimum  node  transmission  range  that  enables  two  nodes  have  direct   communication   and   thus   be   considered   as   neighboring   nodes.   The   software   environment  assures   that   for   every   individual   network  node   there   is   at   least   one   routing  path   from   this   node  towards  the  sink.  

In  every  single  simulation  test,  several  different  topologies  are  created.  For  every  topology,  a  number   of   simulation   runs   is   applied.   During   simulation   tests,   events   are   created   in   arbitrary  positions  and  random  time  instances  and  each  event  range,   i.e.,  a  cyclical  area  that  surrounds  the  event  position,  was  supposed  to  be  varying.  Each  node,  located  within  an  event  range,  is  supposed  to   sense   the  occurrence  of   this  event  and  has   to   transmit   corresponding  event  notifications   to   the  sink.  Since  multiple  nodes  may  serve  the  same  event,  the  sink  may  receive  multiple  notifications  for  the  same  event.  Each  event  notification  is  accompanied  by  a  corresponding  event  message  header.  This  event  header  contains  a  time  stamp  that  denotes  the  time  of  its  generation.  

As   the   event   range   increases   from   an   initial   value   to   a  maximum   value,   the   number   of   the  nodes   that   are   located   within   the   area   where   the   specific   event   takes   place   increases   too.  Subsequently  the  number  of  the  notification  data  sent  to  the  sink  also  arises.  

The   initial  energy  of  every   individual  node  is  considered  to  range  between  a  high  and  a   low  limit  value,  which  are  user  defined.  This  condition  has  been  set   in  order  to  conform  to  the  fact   in  real   WSNs   applications   the   network   nodes   have   different   energy   reserves.   Additionally,   the  dissimilarity  in  the  energy  levels  of  the  sensor  nodes  allows  simulation  tests  to  demonstrate  better  the  ability  of  COALA  protocol  to  achieve  energy  efficient  performance.  

The  configuration  parameters  along  with  their  values  are  summarized  in  Table  2.  

Table  2.  Simulation  Parameters.  

Parameter   Value  Topology  size   1000  m  ×  1000  m  

Number  of  nodes   100  Number  of  different  topologies   100  

Number  of  simulation  runs  for  the  same  topology   30  Node  transmission  range   50  m  

Node  sensing  range   50  m  Node  buffer  size   10  packets  Initial  node  energy   1.5  J–2.5  J  

Event  range   100  m–400  m  Event  range  increase  step   100  m  

4.2.  Presentation  and  Appraisal  of  Simulation  Results  

The   first   set   of   simulation   tests   performed,   evaluate   how   the   increase   in   the   rate   of   data  transmission  influences  the  average  time  it  takes  for  an  event  notification,  sent  by  a  network  node  that  senses  the  specific  event,  to  reach  the  sink.  The  corresponding  simulation  results  are  illustrated  in  Figure  3.  In  these  tests,  all  weighting  factors  are  considered  to  be  equal  to  1.  

As   it   can   be   seen   in   Figure   3,   in  COALA  protocol   the   average  packet   transmission   time  not  only  is  less  than  that  in  DAlPaS  Hard,  but  it  is  also  more  robust  against  the  gradual  growth  of  data  transmission  rate.  

Page 10: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     10  of  15  

 Figure  3.  Average  Packet  Transmission  Time  vs.  Traffic  Load  Rate.  

Next,   the   performance   of   COALA   protocol   is   investigated   relatively   with   how   the   data  transmission  is  affected  by  the  progressive  rise  of  the  of  data  traffic.  This  correlation  is  depicted  in  Figures   4   and   5,   through   the   graphical   representation   of   the   overall   number   of   the   data   packets  received  and  the  percentage  of  the  data  packets  lost  respectively.  

The   examination   of   both   Figures   4   and   5   makes   evident   that   COALA   protocol   not   only  achieves  lower  data  losses  but  also  resists  more  against  the  progressive  increase  of  the  traffic  load.  

The   next   performance   metric   evaluated   is   the   overall   energy   consumption   of   the   network.  Figure  6,  illustrates  in  what  way  the  rise  of  traffic  load  increases  the  network  energy  dissipation.  

 Figure  4.  Number  of  Packets  Received  vs.  Traffic  Load  Rate.  

Page 11: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     11  of  15  

 Figure  5.  Average  Lost  Packets  Ratio  vs.  Traffic  Load  Rate.  

 Figure  6.  Network  Energy  Dissipation  vs.  Traffic  Load  Rate.  

The  examination  of  Figure  6  demonstrates  that  COALA  protocol  outperforms  DAlPaS  Hard  in  energy  efficiency  increasingly  as  the  traffic  load  raises.  

Finally,  a  set  of  simulation  tests  were  performed  in  order  to  examine  how  the  variation  of  the  weighting   factors   of   the   accumulated   cost   of   nodes   deviates   the   results   of   COALA   protocol  utilization.  As  an  example,  the  way  by  which  the  variation  of  two  weighting  factors,  i.e.,    𝑤!   and    𝑤!"   influences   the   average   packet   transmission   time   and   the   number   of   lost   packages   was  investigated.  

Specifically,  the  influence  of  the    𝑤!   variation  in  the  two  aforementioned  metrics  is  depicted  in  Figures  7  and  8  correspondingly.  

Page 12: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     12  of  15  

 Figure  7.  Average  Packet  Transmission  Time  vs.  Traffic  Load  Rate  for  variable    𝑤!.  

 Figure  8.  Number  of  Packets  Lost  vs.  Traffic  Load  Rate  for  variable    𝑤!.  

The   increase   of    𝑤!     makes   the   distance   criterion   have   key   priority  within   the   accumulated  cost   function.  The  examination  of  Figures  7  and  8  validates   that   this   increase  accelerates   the  data  transmission  but  deteriorates   its  quality.  This   is  because  data  are  preferred   to  be   relayed   through  the   closest   neighboring  nodes   although   these  nodes  may  have  more   traffic   load   than  other  more  distant  neighbors.  

Similarly,  the  influence  of  the    𝑤!"   variation  in  the  two  aforementioned  metrics  is  depicted  in  Figures  9  and  10  correspondingly.  

Page 13: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     13  of  15  

 Figure  9.  Average  Packet  Transmission  Time  vs.  Traffic  Load  Rate  for  variable    𝑤!".  

 Figure  10.  Number  of  Packets  Lost  vs.  Traffic  Load  Rate  for  variable    𝑤!".  

The   increase  of    𝑤!"     makes   the  buffer  occupancy  criterion  have  primary  priority  within   the  accumulated   cost   function.   The   examination   of   Figures   9   and   10   confirms   that   this   increase  improves  the  reliability  of  the  data  transmission  at  the  expense  of  the  throughput.  This  is  because  data  are  preferred  to  be  relayed  through  nodes  that  have  less  traffic  load  although  this  may  involve  longer  paths.  

5.  Conclusions  

In  this  research  article,  a  novel  lightweight  scheme  named  COALA,  which  aims  at  preventing  the  diffusion   of   imminent   congestion   in  WSNs   through   alternative   path   routing  was   introduced.  The   innovation   of   this   proposed   scheme   lies   in   the   fact   that   it   takes   into   consideration   a   certain  

Page 14: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     14  of  15  

number  of  both   invariant  and  variable   factors   that  affect   the  probability  of  congestion  occurrence  along  with  other  crucial  factors  like  energy  efficiency.  

The  efficacy  of  COALA  protocol  was  evaluated   through  simulation   tests   in  comparison  with  an  advanced  scheme  of  this  kind,  named  DAlPaS  Hard.  The  first  comparative  advantage  of  COALA  is   that   it   incorporates   the   use   of   the   so-­‐‑called   vicinity   index   during   the   initial   determination   of  routing  paths,  which  makes  nodes  having  a  lot  of  possible  data  receivers  and  few  data  providers  be  favored.   Therefore,   traffic   load   may   be   distributed   in   a   more   balanced   manner.   Additionally,  COALA   introduces   a   multivariable   cost   function,   which   takes   into   consideration   not   only   the  metrics   that   DAlPaS   Hard   suggests,   but   also   the   popularity   index,   the   vicinity   index   and   the  geographical  distance  of  this  node.  Thus,  congestion  caused  due  to  upper-­‐‑level  neighboring  nodes  that   are   repetitively   busy,   or   have  many   probable   data   suppliers   and   few   data   recipients   or   are  located   at  more  distant   locations,   can   be   avoided.  As   a   result,  COALA  achieves   the   reduction   of  transmission  delays,  lost  packets  rates,  and  energy  dissipation.  

Additionally,   it   was   shown   that   the   simple   yet   effective   algorithm   of   COALA   is   able   to  accommodate   to   the   performance   criteria   of   each   individual   application   through   the   appropriate  adjustment  of  the  weighting  of  its  main  cost  function.  

The   authors   of   this   article   intend   in   future   research   work   to   enhance   the   herein-­‐‑proposed  protocol  by  either   incorporating  criteria  based  on  well-­‐‑known  algorithms   for  energy  efficiency   [20–22],   QoS   [23]   and   security   [24].   The   convergence   with   game   theoretic   approaches   [25],   and   the  adaption   of   the   proposed   algorithm   to   standards   for   IPv6   routing   in   LLNs   [26]   are   also   under  consideration.  

Acknowledgments:  A  part  of  the  cost  for  the  publishing  of  this  article  has  been  granted  by  the  University  of  Surrey,  and  the  London  South  Bank  University.  

Author  Contributions:  Dionisis   Kandris   conceived   the   proposed   protocol,   designed   the   protocol   algorithm  and  wrote  the  manuscript;  George  Tselikis  contributed  to  the  refinement  of  the  protocol  algorithm,  developed  the   simulation   environment   and   performed   the   simulation   tests;   Eleftherios   Anastasiadis,   Emmanouil  Panaousis  and  Tasos  Dagiuklas  contributed  to  the  refinement  of  the  protocol  algorithm  and  the  interpretation  of  the  simulation  results.  

Conflicts  of  Interest:  The  authors  declare  no  conflict  of  interest.  

References  

1. Sergiou,   C.;   Antoniou,   P.;   Vassiliou,   V.   A   comprehensive   survey   of   congestion   control   protocols   in  wireless  sensor  networks.  IEEE  Commun.  Surv.  Tutor.  2014,  16,  1839–1859.  

2. Ghaffari,  A.  Congestion  control  mechanisms  in  wireless  sensor  networks:  A  survey.  J.  Netw.  Comput.  Appl.  2015,  52,  101–115.  

3. Ploumis,  S.E.;  Sgora,  A.;  Kandris,  D.;  Vergados,  D.D.  Congestion  Avoidance  in  Wireless  Sensor  Networks:  A  Survey.  In  Proceedings  of  the  16th  Panhellenic  Conference  on  Informatics  (PCI  2012),  Piraeus,  Greece,  5–7  October,  2012;  pp.  234–239.  

4. Sumathi,   K.;   Venkatesan,   M.   A   Survey   on   Congestion   Control   in   Wireless   Sensor   Networks.   Int.   J.  Comput.  Appl.  2016,  147,  6–11.  

5. Dashkova,   E.;   Gurtov,   A.   Survey   on   congestion   control   mechanisms   for   wireless   sensor   networks.   In  Internet   of   Things,   Smart   Spaces,   and  Next   Generation  Networking;   Springer:   Berlin/Heidelberg,   Germany,  2012;  pp.  75–85.  

6. Kaur,  J.;  Grewal,  R.;  Saini,  K.S.  A  survey  on  recent  congestion  control  schemes  in  wireless  sensor  network.  In  Proceedings  of  the  2015  IEEE  International  Advance  Computing  Conference  (IACC),  Bangalore,  India,  11–13  June  2015;  pp.  387–392.    

7. Rathod,  H.M.;  Buddhadev,  B.V.  Comparative  study  of  congestion  control  techniques  for  wireless  sensor  network.   In   Proceedings   of   the   2011   Nirma   University   International   Conference   on   Engineering  (NUiCONE),  Gujarat,  India,  8–10  December  2011;  pp.  1–5.  

8. Wan,   C.;   Eisenman,   S.B.;   Campbell,   A.T.   CODA:   Congestion   Detection   and   Avoidance   in   Sensor  Networks,  In  Proceedings  of  the  1st  International  Conference  on  Embedded  Networked  Sensor  Systems  (SenSys  ’03),  Los  Angeles,  CA,  USA,  5–7  November  2003;  pp.  266–279.  

Page 15: Article! COALA:AProtocolfortheAvoidanceandAlleviation ... · Sensors!2017,17,2502!! 6!of!15!2017,17,2502!! 6!of!15!

Sensors  2017,  17,  2502     15  of  15  

9. Ahmad,  M.;  Turgut,  D.  Congestion  Avoidance  and  Fairness  in  Wireless  Sensor  Networks.  In  Proceedings  of   the   2008   IEEE   Global   Telecommunications   Conference   (GLOBECOM),   New   Orleans,   LA,   USA,   30  November–4  December  2008;  pp.  1–6.  

10. Kandris,  D.;   Vergados,  D.J.;   Vergados,  D.D.;   Tzes,  A.  A  Routing   Scheme   for  Congestion  Avoidance   in  Wireless   Sensor  Networks.   In   Proceedings   of   the   6th  Annual   IEEE  Conference   on  Automation   Science  and  Engineering  (CASE  2010),  Toronto,  ON,  Canada,  21–24  August  2010;  pp.  497–502.  

11. Sergiou,  C.;  Vassiliou,  V.  DAlPaS:  A  performance  aware  congestion  control  algorithm  in  Wireless  Sensor  Networks.  In  Proceedings  of  the  18th  IEEE  International  Conference  on  Telecommunications  (ICT),  Ayia  Napa,  Cyprus,  8–11  May  2011;  pp.  167–173.  

12. Huang,   J.-­‐‑M.;   Li,   C.-­‐‑Y.;   Chen,   K.-­‐‑H.   TALONet:   A   power-­‐‑efficient   grid-­‐‑based   congestion   avoidance  scheme  using  multi-­‐‑detouring  technique  in  wireless  sensor  networks.  In  Proceedings  of  the  IEEE  Wireless  Telecommunication  Symposium  (WTS  2009),  Prague,  Czech  Republic,  22–24  April  2009;  pp.  1–6.  

13. Sergiou,  C.;  Vassiliou,  V.;  Pitsillides,  A.  Reliable  data  transmission  in  event-­‐‑based  sensor  networks  during  overload  situation.  In  the  Proceedings  of  the  3rd  International  Conference  on  Wireless  Internet,  WICON  2007,  Austin,  TX,  USA,  22–24  October  2007;  pp.  1–8.  

14. He,  T.;  Ren,  F.;  Lin,  C.;  Das,  S.  Alleviating  Congestion  Using  Traffic-­‐‑Aware  Dynamic  Routing  in  Wireless  Sensor  Networks.  In  Proceedings  of  the  5th  Annual  IEEE  Communications  Society  Conference  on  Sensor,  Mesh   and  Ad  Hoc  Communications   and  Networks   (SECON   ’08),   San   Francisco,  CA,  USA,   16–20   June  2008;  pp.  233–241.  

15. Tao,  L.Q.;  Yu,  F.Q.  ECODA:  Enhanced  congestion  detection  and  avoidance  for  multiple  class  of  traffic  in  sensor  networks.  IEEE  Trans.  Consum.  Electron.  2010,  56,  1387–1394.  

16. Kang,   J.;   Zhang,   Y.;  Nath,   B.   TARA:   Topology-­‐‑Aware  Resource  Adaptation   to  Alleviate  Congestion   in  Sensor  Networks.  IEEE  Trans.  Parallel  Distrib.  Syst.  2007,  18,  919–931.  

17. Jan,  M.A.;  Nanda,   P.;  He,   X.;   Liu,   R.P.   PASCCC:   Priority-­‐‑based   application-­‐‑specific   congestion   control  clustering  protocol.  Comput.  Netw.  2014,  74,  92–102.    

18. Tselikis,  G.  Cross-­‐‑Layer  Protocol  Design  of  a  3G-­‐‑RNC  Simulator.  In  Proceedings  of  the  Fourth  Advanced  International  Conference  on  Telecommunications  AICT’08,  Athens,  Greece,  8–13  June  2008;  pp.  168–172.  

19. Tselikis,  G.  Implementation  of  the  Radio  Network  Controller  Signaling  Protocol  Stack  using  Cross  Layer  Design.  Int.  J.  Electron.  Inf.  Syst.  2010,  12,  13–22.  

20. Kandris,   D;   Tsioumas,   P.;   Tzes,   A.;  Nikolakopoulos,   G.;   Vergados,   D.D.   Power   Conservation   Through  Energy  Efficient  Routing  in  Wireless  Sensor  Networks.  Sensors  2009,  9,  7320–7342.  

21. Nikolidakis,  S.A.;  Kandris,  D.;  Vergados,  D.D.;  Douligeris,  C.  Energy  efficient  routing  in  wireless  sensor  networks  through  balanced  clustering.  Algorithms  2013,  6,  29–42.  

22. Politis,   I.;   Tsagkaropoulos,   M.;   Dagiuklas,   T.;   Kotsopoulos,   S.   Power   efficient   video   multipath  transmission  over  wireless  multimedia  sensor  networks.  Mob.  Netw.  Appl.  2008,  13,  274–284.  

23. Kandris,  D.;  Tsagkaropoulos,  M.;  Politis,  I.;  Tzes,  A.;  Kotsopoulos,  S.  Energy  efficient  and  perceived  QoS  aware  video  routing  over  wireless  multimedia  sensor  networks.  Ad  Hoc  Netw.  2011,  9,  591–607.  

24. Panaousis,  E.;  Karapistoli,  E;  Hadeer,E.;  Alpcan,  T.;  Khuzani,  MHR.;  Economides,  A.A.  Game   theoretic  path   selection   to   support   security   in   device-­‐‑to-­‐‑device   communications.  Ad  Hoc   Netw.   2017,   56,   28–42,  doi:10.1016/j.adhoc.2016.11.008.  

25. Anastasiadis,  E.;  Deng,  X.;  Krysta,  P.;  Li,  M.;  Qiao,  H.;  Zhang,  J.  New  results  for  network  pollution  games.  In   Proceedings   of   the   22nd   International   Computing   and   Combinatorics   Conference   COCOON   2016,    Ho  Chi  Minh  City,  Vietnam,  2–4  August  2016;  pp.  39–51.  

26. Weisheng,   T.;   Xiaoyuan,   M.;   Jun,   H.;   Jianming,   W.   Toward   Improved   RPL:   A   Congestion   Avoidance  Multipath   Routing   Protocol   with   Time   Factor   for   Wireless   Sensor   Networks.   J.   Sens.   2016,   2016,  doi:10.1155/2016/8128651.  

©   2017   by   the   authors.   Submitted   for   possible   open   access   publication   under   the    terms   and   conditions   of   the   Creative   Commons   Attribution   (CC   BY)   license  (http://creativecommons.org/licenses/by/4.0/).